Sparse grids in their basic form requires a maximal resolution to be specified. In reality, however, problems are not likely to be isotropic i.e. require the same resolution in all dimensions nor is it likely that the right resolution is known apriori. Adaptive sparse grids automatically search for the best resolution in each dimension and builds a model of the data. This project will study techniques, implementations and merits of such adaptive schemes for use in predictive modelling using sparse grids.